Chromatic

On zeros of the characteristic of representable matroids of bounded -width.

Carolyn Chun § Rhiannon Hall § Criel Merino‡ Steven D. Noble§

§Departament of Mathematical Sciences Brunel University

‡Instituto de Matem´aticasUNAM, sede Oaxaca

Modern Techniques in Discrete Optimization: , Algorithms and Applications, BIRS Oaxaca, 1-6 November 2015

Criel Merino On zeros of the characteristic polynomial of representable matroids of bounded tree-width. Chromatic Matroids Chromatic polynomial

The chromatic polynomial (for planar graphs) was first defined by Birkhoff in 1912 in an attempt to find an algebraic proof of the then open four-colour problem. The chromatic polynomial χ(G; λ) of a graph G counts, for every positive integer λ, the number of proper colouring of the vertices of G with λ colours. One can easily check that, if e is an edge of G, then

P(G, q) = P(G \ e, q) − P(G/e, q).

Criel Merino On zeros of the characteristic polynomial of representable matroids of bounded tree-width. Chromatic Matroids Chromatic polynomial

The following theorem is due to G. D. Birkhoff (and independently by H. Whitney1932) Theorem If G = (V , E) is a graph, then X χ(G; λ) = λκ(E) (−1)|A|λr(E)−r(A). (1) A⊆E

Criel Merino On zeros of the characteristic polynomial of representable matroids of bounded tree-width. Chromatic Matroids Chromatic polynomial

Since the chromatic polynomial can be evaluated at real and complex values, Birkhoff hoped to use analytic methods to prove the 4-colour conjecture. In fact, the four-colour Theorem is equivalent to stating that χ(G; 4) > 0 for all planar graphs G. Such an analytic proof was never found, but Birkhoff and Lewis in 1946 did show that if G is planar then χ(G; λ) > 0 for all real λ ≥ 5. They conjectured that 5 could be changed to 4, and almost 70 years later their conjecture is still open. Conjecture (Birkhoff-Lewis, 1946) If G is a , then χ(G; λ) > 0 for all real λ ≥ 4.

Criel Merino On zeros of the characteristic polynomial of representable matroids of bounded tree-width. Chromatic Matroids Chromatic polynomial

What about largest real roots? A family of graphs G has an upper root-free interval if there is a real number a such that no graph in G has a real root larger than a. The class of all graphs does not have an upper root-free interval (since you can get arbitrarily large chromatic roots from cliques).

Criel Merino On zeros of the characteristic polynomial of representable matroids of bounded tree-width. Chromatic Matroids Chromatic polynomial

This is different for a proper minor-closed class of graphs. Theorem (Mader ’67) For every k ∈ N there exists an integer f (k) such that any graph with minimum degree at least f (k) has a Kk -minor.

It follows that, for every proper minor-closed family of graphs G, there exists a smallest integer d = d(G) such that every graph in G has a vertex of degree at most d. Woodall and Thomassen proved the next result independently. Theorem (Thomassen ’97, Woodall ’97) If G is a proper minor-closed family of graphs, then (d(G), ∞) is an upper root-free interval for G.

Criel Merino On zeros of the characteristic polynomial of representable matroids of bounded tree-width. The idea of tree-width in graphs:

To say how “tree-like” the structure of a graph is.

tree decomposition

The width of a tree decomposition is the highest number of graph vertices contained in a node of the tree, minus one. The idea was extended to matroids by Hlineny and Whittle The idea was extended to matroids by Hlineny and Whittle

tree decomposition

Tree decomposition: • The nodes of � are “buckets”. • Each element of � goes in one bucket (partitioning �(�)). Buckets can be empty or contain many elements. Tree-Width for Matroids

• Given a tree decomposition, each node gets a weight called its node-width.

• The width of a tree decomposition is the highest value of its node-widths.

• The tree-width of a is the minimum width of all its tree decompositions. Tree-Width for Matroids

• Given a tree decomposition, each node gets a weight called its node-width.

• The width of a tree decomposition is the highest value of its node-widths.

• The tree-width of a matroid is the minimum width of all its tree decompositions.

• How do we measure node-width? The width of a node � of � is

� � = � � − ∑ � � − � � � − �

= ∑� � � − � − � − 1 � � . The width of a node � of � is

� � = � � − ∑ � � − � � � − �

= ∑� � � − � − � − 1 � � .

What does this value mean geometrically? Brief Digression: A note on �-separations in representable matroids.

Let � be a matroid representable over ��(�). Then we can embed � in the projective space ��(� � − 1,�). Let (�, �) be a �-separation of �.

� �� � ∩ �� � � “guts” of (�, �)

plonk in points to get projective space here

Then we can add points of ��(� � − 1,�) to � into the guts of (�, �) to obtain matroid �′ and �-separation (�, � ) such that the guts of (�, �) is the projective space ��(� − 2, �).

The width of a node � of � is

� � = � � − ∑ � � − � � � − �

= ∑� � � − � − � − 1 � � .

What does this value mean geometrically? If � is representable, then the width of node � is

� � = � � ∪ � ∪ � ∪ ⋯ ∪ � . If � is representable, then the width of node � is

� � = � � ∪ � ∪ � ∪ ⋯ ∪ � .

Note: If � is a leaf node of �, and � is the set of matroid elements in the �- bucket, then � � = � � . Example: Tree decompositions for �, and �,

� 1 �, � 2

� � � � � � � � 1 2 3 3 3 3 2 1 node widths

� 1 �, � 2

� � � � � � � � 1 2 3 6 4 3 2 1 node widths Example: Tree decompositions for a daisy

� 3 3 �

ø � 3 5 � 3 node widths � 3 Example: Tree decompositions for a daisy

� 3 3 �

ø � 3 5

� 3 � 3 

� 3 � 3

ø 3

3 � node widths ø 3 ø 3 � 3  � 3 The Chromatic Polynomial for Matroids

For a matroid M,

� �, � = (−1)||� (). ⊆

The familiar deletion-contraction rules apply: • If � is not a loop or coloop then � �, � = � �\�, � − � �/�, � , • If � is a loop then � �, � = 0, • If � is a coloop then � �, � = � − 1 � �\�, � . Chromatic Matroids The Characteristic polynomial

For example,the characteristic polynomial of Um,n, 0 < m ≤ n, is Pm−1 k n m−k χUm,n (λ) = k=0 (−1) k (λ − 1).

For PG(r − 1, q), whose lattice of flats is isomorphic to the lattice of subspaces of the r-dimensional vector space over GF (q), has characteristic polynomial 2 r−1 χPG(r−1,q)(λ) = (λ − 1)(λ − q)(λ − q ) ··· (λ − q ).

Criel Merino On zeros of the characteristic polynomial of representable matroids of bounded tree-width. Chromatic Matroids generalized parallel connection

Given two matroids M1 = (E1, r1) y M2 = (E2, r2) we define the generalized parallel connection, PN (M1, M2), as the matroid over E1 ∪ E2 whose flats are the sets X de E1 ∪ E2 such that X ∩ E1 is a flat in M1 and X ∩ E2 is a flat in M2 and where ∼ ∼ N = M1|T = M2|T and T = E1 ∩ E2. Then,

χM1 (λ)χM2 (λ) χPN (M1,M2)(λ) = . (3) χN (λ)

Criel Merino On zeros of the characteristic polynomial of representable matroids of bounded tree-width. Theorem (Chun, H, Merino, Noble): Let � be a matroid representable over ��(�). If � has tree-width at most �, then there exists � ∈ ℝ such that � �, � > 0 for all � > �. Proof Outline:

For induction, lexicographically order all �� � -representable matroids by (rank, |� � |, something).

We make sure � is large enough that �� 0, � ,�� 1, � ,… , �� � − 1,� satisfy the theorem. Induction: Let � be �� � -representable with tree-width ≤ �. Assume all �� � -representable matroids of tree-width ≤ � that occur before � in the LEX ordering satisfy the theorem. Let � be an optimal tree decomposition of � and let (�, �) be a �′-separation of � displayed by � (note �′ ≤ �): Let {�, �, … , �} be the elements of �� �(�) − 1, � in the guts of �, � .

Let � denote the matroid � extended by �.

Lemma: �,,…, has the same tree-width as � (construct the same tree decomposition with �, �,… , � in bucket �or �).

Lemma: For any matroid � and any element � ∈ �(�), ��(�/e) ≤ ��(�). Induction:

,,…, We need to know that �(� , �) is positive for all � ≥ �. Lemma: If the guts is a modular flat then

� (�|�)… , � �((�|�)… , �) � �… , � = �({�, �, … , �}, �) � (�|�)… , � �((�|�)… , �) � �… , � = �({�, �, … , �}, �)

Rank lower than � so the induction works … … except in the case where � has no �′-separations for any � < � � . In which case, extend M to �� �(�) − 1, � as on the previous slide.